The Global Phosphorylation Landscape of SARS-CoV-2 Infection
A time-resolved phosphoproteomic survey of SARS-CoV-2-infected cells showing that infection acts mainly through signalling rather than protein abundance, activating casein kinase II and the p38 cascade while shutting down mitotic kinases, and converting that kinase profile into inhibitors with antiviral activity.
- Senior authors
- Nevan J. Krogan
- Correspondence
- Robert Grosse; Adolfo García-Sastre; Marco Vignuzzi; Jeffery R. Johnson; Kevan M. Shokat; Danielle L. Swaney; Pedro Beltrao; Nevan J. Krogan
Research areas & themes
Citation
Bouhaddou M, Memon D, Meyer B, White KM, Rezelj VV, Correa Marrero M, et al. The Global Phosphorylation Landscape of SARS-CoV-2 Infection. Cell, 2020, volume 182, issue 3, pages 685-712.e19.
DOI 10.1016/j.cell.2020.06.034. PMID 32645325. PMCID PMC7321036.
One-sentence contribution
A time-resolved phosphoproteomic survey of SARS-CoV-2-infected cells showing that infection acts mainly through signalling rather than protein abundance, activating casein kinase II and the p38 cascade while shutting down mitotic kinases, and converting that kinase profile into inhibitors with antiviral activity.
Executive summary
Early in the COVID-19 pandemic the options against SARS-CoV-2 were limited to remdesivir and supportive care, which made host-directed approaches attractive. This study asked how the virus rewires host signalling, on the reasoning that kinases are both readable from phosphorylation data and druggable. Vero E6 cells were infected and harvested in biological triplicate at six time points over 24 hours, and each sample was split for measurement of protein abundance and of phosphorylation by mass spectrometry, with sites mapped to human orthologs. Regulation was overwhelmingly at the level of phosphorylation rather than abundance, and almost no significantly changed phosphorylation site had a matching abundance change, which the authors read as evidence that signalling rather than transcription is the primary host response over this interval. Kinase activities were inferred for 97 of the 518 human kinases. Casein kinase II and multiple p38 pathway members rose, while cyclin-dependent and mitotic kinases fell, and flow cytometry confirmed arrest between S and G2. Imaging of infected human Caco-2 cells showed a marked increase in long branched filopodia bearing viral M protein, with casein kinase II partially co-localizing with N protein along them, and electron microscopy showed assembled particles on and apparently budding from these protrusions. Inhibition of p38 reduced inflammatory cytokines and viral replication. Mapping regulated kinases onto known inhibitors gave 87 candidates, of which 68 were tested at two institutions in two cell lines, with antiviral activity for inhibitors of casein kinase II, p38, AXL, PIKFYVE and cyclin-dependent kinases.
Scientific context
SARS-CoV-2 is an enveloped positive-sense RNA betacoronavirus closely related to SARS-CoV, entering cells through ACE2 with spike priming by TMPRSS2, translating ORF1a and ORF1ab into non-structural proteins that assemble the replication and transcription complex within remodelled endoplasmic reticulum, and assembling at the endoplasmic reticulum to Golgi intermediate compartment before release. At the time of writing, clinical management was largely supportive and remdesivir was the only agent with emergency authorisation. A virus to human protein interaction map for SARS-CoV-2 had recently been published by the same broad consortium, identifying 332 human proteins interacting with 27 viral proteins, which supplied a physical map but not a functional one. What remained unresolved was how infection alters the activity state of host signalling, which matters because the functional consequences of many phosphorylation events are annotated, because kinase activities can be inferred from substrate phosphorylation, and because kinases are a well-populated drug target class.
Central question
How does SARS-CoV-2 infection rewire host protein phosphorylation and kinase activity over the course of infection, and can the resulting activity profile be mapped onto existing kinase inhibitors to identify compounds with antiviral efficacy.
Experimental strategy
The design rests on treating phosphorylation as the readout with the right time resolution for an acute infection. Vero E6 cells were used for the discovery proteomics because of their permissiveness, and each sample was divided so that protein abundance and phosphorylation were measured from the same material, which allows the two layers to be compared directly rather than inferred against each other. Six time points spanning entry, replication and egress were used so that changes could be assigned to a phase of the life cycle, with biological triplicates and mock controls at both the start and end of the interval. Because the host is African green monkey, sequences were aligned to human and sites mapped to human orthologs, which makes the dataset interpretable against human annotation at the cost of introducing an orthology mapping step. Kinase activity was then inferred from the coordinated regulation of annotated substrates rather than measured directly, and the inferred profile was compared against a compendium of published phosphoproteomic conditions to place the infected state relative to known perturbations. Findings from that inference layer were followed with orthogonal assays in human cells, including immunoblotting for activating phosphorylations, imaging and electron microscopy, DNA content analysis, cytokine measurement by transcript and by protein, and siRNA knockdown. Pharmacological testing was deliberately spread across two institutions and two cell lines with overlapping and distinct inhibitor target profiles, which is what allows target attribution rather than compound-level observation alone.
Key findings
-
Phosphorylation, not protein abundance, is the dominant layer of the host response over 24 hours. High-quality quantification covered 4,624 human-orthologous phosphorylation sites and 3,036 proteins. The number of significantly regulated phosphorylation sites rose across the time course, while few proteins changed in abundance, and most that increased were viral (Fig. 1E to 1H). For nearly every significantly changed phosphorylation site there was no corresponding abundance change (Fig. 1J). The authors interpret this as evidence that signalling rather than transcriptional regulation is the primary host response in this window.
-
Most host proteins that changed in abundance decreased, with Gene Ontology enrichment for platelet regulation among the downregulated set, including APOH, CD9, TSPAN14, AHSG, SERPINA1 and A2M (Fig. 1I). The suggestion that this may relate to coagulation and stroke in patients is explicitly framed by the authors as a possibility rather than a demonstrated link.
-
Forty-nine phosphorylation sites across seven viral proteins were catalogued, 25 detected here and the remainder from a published dataset, with sites mapped for conservation, secondary structure and predicted interface status (Fig. 2A and 2B). Casein kinase II, cyclin-dependent kinase and protein kinase C families were the top predicted upstream kinases by sequence (Fig. 2C). The proposals that a C-terminal negative charge cluster in M protein is functional, that phosphorylation alters the surface charge of the N protein acidic wrist and thereby modulates RNA binding allosterically, and that the RS-rich region behaves as in SARS-CoV, are all presented by the authors as hypotheses.
-
Forty of the 332 host proteins in the published virus to human interaction map were significantly differentially phosphorylated during infection (Fig. 3). Interpretations offered for individual cases, including decreased LARP1 phosphorylation favouring inhibition of ribosomal protein synthesis and increased NUP98 S888 phosphorylation possibly blocking host mRNA export, are author interpretation drawn from published functions of those sites rather than tested here.
-
Regulated sites fell into five temporal clusters that the authors align with entry, replication and egress, with early clusters enriched for mRNA processing, cell cycle and apoptosis and later clusters enriched for RNA processing and DNA replication (Fig. 4A). Activity regulation was estimated for 97 kinases, strongest at 0 to 2 hours and at 24 hours. Predicted activation covered p38 pathway members including MAPK12, casein kinase II subunits, CAMK2G and PRKG1 and PRKG2, while predicted downregulation covered CDK1, CDK2, CDK5, AURKA, PRKACA, AKT1 and AKT2, MAPK1 and MAPK3, PIM1 and PAK1 (Fig. 4B).
-
Comparison against a compendium of published phosphoproteomic conditions placed early and late infection near states induced by inhibition of mTOR, ERK, AKT and EGFR, and the middle of the time course near inhibition of PI3K, p70S6K and ROCK, with several time points resembling an S to G2 state and anticorrelating with mitosis (Fig. 4D). Complex-level analysis showed changes in spliceosome, proteasome and chromatin remodelling complexes (Fig. 4E).
-
Cytoskeletal signalling is reorganised. PAK1 and PAK2 target sites in vimentin and stathmin were downregulated, while well-characterized casein kinase II target sites including CTNNA1 S641 and MYH9 S1943 rose (Fig. 5A and 5B). In SARS-CoV-2-infected human Caco-2 cells, M protein clusters localized along the shafts and tips of actin-rich filopodia, and infection increased filopodial number, length and branching relative to mock (Fig. 5C and 5D). Casein kinase II was present along these protrusions and partially co-localized with N protein (Fig. 5E and 5F). Scanning and transmission electron microscopy showed assembled particles along the filopodia with instances appearing to bud from them (Fig. 5G and 5H). Phosphoproteomics of Vero E6 cells overexpressing N protein alone showed significantly increased casein kinase II activity (Fig. S3E). The proposal that N protein allosterically controls casein kinase II activity to regulate the cytoskeleton is stated by the authors as a hypothesis.
-
The p38 cascade is activated during infection. Inferred activation of MAP2K3, MAP2K6, MAPK12, MAPKAPK2 and MAPKAPK3 was confirmed by immunoblotting for phospho-p38 T180 and Y182, phospho-MK2 T334 and phospho-CREB and phospho-ATF1 in ACE2-expressing A549 human lung cells (Fig. 6B and 6C). Substrate sites including NELFE, HSPB1 and STAT1 rose late in the time course (Fig. 6D). Transcription factor activity inferred from published expression data in A549, Calu-3 and primary human bronchial epithelial cells showed p38-regulated factors among the most activated (Fig. 6E).
-
The p38 inhibitor SB203580 reduced IL-6 and TNF-alpha transcripts dose dependently in infected ACE2-A549 cells, and multiplexed ELISA confirmed reductions in IL-6, CXCL8, CCL20 and CCL2 protein (Fig. 6F, Fig. S4A and S4B). The same treatment reduced SARS-CoV-2 subgenomic mRNA without major toxicity. The authors state explicitly that because replication is also inhibited, the contributions of p38 activity and of viral presence to cytokine production cannot be deconvolved in this experiment.
-
Infection arrests the cell cycle. Phosphoproteomic profiles correlated with cells arrested at the S to G2 transition and anticorrelated with mitotic cells (Fig. 6G). CDK2 T14 and Y15 phosphorylation rose at 2 hours then declined, as did H2AX S140 (Fig. 6H). DNA content analysis at 24 hours showed a significant increase in cells in S phase and at G2 to M and a decrease in G0 to G1 (Fig. 6I). Whether the arrest is caused by p38 activity is raised as a possible mechanistic link from prior literature and is not tested.
-
Mapping regulated kinase activities to inhibitors yielded 87 drugs and compounds, 10 approved by the US Food and Drug Administration, 53 in clinical testing and 24 preclinical, of which 68 were tested for antiviral efficacy and cytotoxicity at Mount Sinai and Institut Pasteur, in Vero E6 and ACE2-A549 cells (Fig. 7A). Remdesivir served as a positive control with a half maximal inhibitory concentration of 1.28 micromolar (Fig. 7B). Antiviral activity was found for the casein kinase II inhibitor silmitasertib at 2.34 micromolar, the AXL inhibitor gilteritinib at 0.807 micromolar, the MAPK11 and MAPK14 inhibitor ralimetinib at 0.873 micromolar, MAPK13-IN-1 at 4.63 micromolar, the MAPK14 inhibitor ARRY-797 at 0.913 micromolar in ACE2-A549 cells, the PIKFYVE inhibitor apilimod below 0.08 micromolar in Vero E6 and at 0.007 micromolar in ACE2-A549, and the CDK inhibitor dinaciclib at 0.127 and 0.032 micromolar in the two lines (Fig. 7C and 7E to 7K).
-
Genetic perturbation supported the pharmacology for the p38 arm. Small interfering RNA knockdown of MAP2K3, MAPK13 and MAPK12 in ACE2-A549 cells significantly decreased viral replication with little or no effect on viability (Fig. 7I).
Mechanistic model
The study does not establish a single mechanism by which SARS-CoV-2 rewires host signalling, and the drug results identify host dependencies without resolving how each one is used by the virus. The authors state plainly that p38 inhibition suppresses cytokine production and impairs replication by a still unknown mechanism.
What the data do constrain is the following. Over 24 hours the host response is enacted through phosphorylation rather than through changes in protein amount, which the paired abundance and phosphorylation measurement supports directly. Kinase activity changes are inferred, not measured, from coordinated substrate regulation, so statements about the activity of individual kinases carry the confidence of that inference except where confirmed by immunoblotting, as was done for the p38 arm. Two activity changes are supported by convergent evidence. Casein kinase II activity rises, which is supported by substrate phosphorylation, by the independent observation that N protein overexpression alone raises inferred casein kinase II activity, by physical interaction between N protein and casein kinase II subunits reported in the companion interaction map, by co-localization at filopodia, and by antiviral activity of silmitasertib. The p38 cascade is activated, supported by substrate phosphorylation, by immunoblotting for activating sites in human cells, by transcription factor activity inference from independent expression datasets, and by concordant pharmacological and siRNA results.
The proposed models beyond those points are the authors' own. That virus-containing filopodia serve egress or cell-to-cell spread is presented as a hypothesis, with the discussion noting the resemblance to Marburg virus more than to vaccinia and stating that further work is needed to determine whether the virus exploits Myosin X motor activity or actin assembly. That N protein allosterically controls casein kinase II is a hypothesis. That S to G2 arrest benefits replication by supplying nucleotides and DNA repair and replication proteins is imported from prior literature on other viruses. The co-localization and imaging data are correlative and do not establish that casein kinase II activity at filopodia is required for particle release.
Conceptual or technical advance
The work supplies a functional layer on top of the physical virus to host interaction map for SARS-CoV-2, and it demonstrates a workflow in which a time-resolved phosphoproteomic profile is converted into inferred kinase activities and then into a ranked set of testable compounds, with the compound set constructed so that overlapping and distinct target profiles allow attribution to targets rather than to molecules. Several specific host dependencies became testable as a result, including casein kinase II, the p38 cascade and its upstream activator AXL, PIKFYVE, and cyclin-dependent kinases. The observation that infection drives formation of long branched filopodia carrying viral protein and apparently budding particles opens a route of egress and spread for examination. The finding that p38 inhibition reduces both inflammatory cytokine output and viral replication distinguishes SARS-CoV-2 from earlier cases the authors cite for SARS-CoV, dengue virus and influenza A virus, where p38 inhibition affected the host response without directly impairing the virus, though the basis of that difference is not established here.
Relationship to the broader research program
The tenOever contribution to this paper is one component of a large multi-institution collaboration. The author contributions statement lists tenOever under work supervision, alongside fifteen others, and a member of his group appears among those who performed infection experiments. The transcription factor activity analysis draws on expression data from A549, Calu-3 and primary human bronchial epithelial cells reported in Blanco-Melo and colleagues 2020, work in which that laboratory was involved, which is the main point of contact between this study and the laboratory's own line of work on the transcriptional host response to SARS-CoV-2.
Category 3 synthesis, visible only across papers. The interferon and STAT1 axis appears here as a phosphoproteomic observation, with STAT1 among the p38 substrate sites rising late in infection, and the same protein is the object of direct mechanistic work in the 2007 Science report on IKKε. That is a shared subject rather than a claimed lineage, and neither paper cites the other. The SARS-CoV-2 hamster work reported by Horiuchi and colleagues in 2021 from the tenOever laboratory addresses immunity and transmission in an animal host, a different level of the same pathogen, and is not connected to this study by shared data or method.
Related publications
- Gordon et al. 2020, companion. The SARS-CoV-2 virus to human protein interaction map from the same consortium supplies the 332 interacting proteins used here and the physical interactions invoked for casein kinase II, Nsp7 with RHOA and Nsp2 with WASHC5.
- Blanco-Melo et al. 2020, methodological foundation for the transcription factor activity analysis, supplying expression data from infected A549, Calu-3 and primary human bronchial epithelial cells.
- Davidson et al. 2020, companion dataset from another group whose viral phosphorylation sites were combined with those detected here.
- Ou et al. 2020, predecessor reporting antiviral capacity for apilimod, which this study extends by placing PIKFYVE within a phosphorylation-regulated context.
- Horiuchi et al. 2021, Science Immunology, from the tenOever laboratory. Shares the pathogen but no data, method or claim with this study, so no substantive relationship is asserted.
Limitations and boundaries
The authors state their own principal limitation, that the discovery proteomics was performed in Vero E6, an African green monkey kidney line rather than a human respiratory cell, with the mitigation that pharmacological testing was repeated in human ACE2-A549 cells and that most drug effects replicated across lines. Several further boundaries follow from the design. Phosphorylation sites were mapped onto human orthologs, which adds an inference step between measurement and annotation. Kinase activities were estimated from substrate regulation for 97 of 518 human kinases, so the analysis is bounded by existing substrate annotation and cannot speak to poorly annotated kinases, and individual activity calls are predictions except where confirmed directly, as the paper itself illustrates for PRKACA, where substrate-based inference predicted decreased activity while an activation loop site increased. The viral phosphorylation site catalogue does not distinguish cleaved from uncleaved viral proteins. Structural interpretations of M protein and N protein phosphorylation are computational predictions without functional testing. The cytokine experiment cannot separate the effect of p38 activity from the effect of reduced viral replication, as stated in the text. The imaging and electron microscopy evidence for filopodial budding is correlative, and no perturbation establishes that these structures are required for egress or spread. Antiviral efficacy is measured in cell culture over 48 hours with cell viability controls, and no animal or clinical efficacy is shown, so the compounds are candidates rather than established therapies. Inhibitor selectivity is a general caveat the paper addresses by testing overlapping target profiles rather than by direct target engagement measurement. The single virus isolate and the single cell type used for discovery mean that variant-dependent and cell-type-dependent aspects of the signalling response are outside the study's scope.
Audience summaries
25 words
Time-resolved phosphoproteomics of infected cells showed SARS-CoV-2 acts mainly by rewiring signalling, activating casein kinase II and p38 while arresting the cell cycle, nominating several antiviral inhibitors.
75 words
SARS-CoV-2-infected cells were sampled at six time points and analyzed for protein amount and phosphorylation. Almost all regulation occurred through phosphorylation. Casein kinase II and the p38 cascade were activated, mitotic kinases shut down, and cells arrested between S and G2. Infected human cells formed long branched filopodia carrying viral protein and budding particles. Inhibitors of casein kinase II, p38, AXL, PIKFYVE and cyclin-dependent kinases showed antiviral activity in culture.
150 words
Kinases are readable from phosphorylation data and well supplied with inhibitors, which makes them a practical entry point for host-directed antivirals. Infected Vero E6 cells were sampled in triplicate at six time points, each sample split for abundance and phosphorylation measurement. Regulation was overwhelmingly post-translational, with almost no phosphorylation change accompanied by an abundance change. Activity was inferred for 97 kinases, with casein kinase II and p38 cascade members rising and cyclin-dependent and mitotic kinases falling, and flow cytometry confirmed arrest between S and G2. In human Caco-2 cells, infection induced long branched filopodia bearing viral M protein, with casein kinase II partially co-localizing with N protein, and electron microscopy showed particles apparently budding from these structures. p38 inhibition reduced both inflammatory cytokines and viral replication. Of 87 mapped compounds, 68 were tested, and inhibitors of casein kinase II, p38, AXL, PIKFYVE and cyclin-dependent kinases showed antiviral activity.
Documented publication relationships
- Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19 — methodological foundation.
- Immune memory from SARS-CoV-2 infection in hamsters provides variant-independent protection but still allows virus transmission — related.
- Immune memory from SARS-CoV-2 infection in hamsters provides variant-independent protection but still allows virus transmission — related.